Transforming Retail: Building an AI Product for Nike
Table of Contents:
- Introduction
- Background and Experience with AI and Machine Learning
- Building a Successful AI Product for Retail
3.1 Finding a Friendly AI Problem
3.2 Designing the AI Product and Implementation
3.3 The Importance of Measurement
- Where To Use AI in the Retail Cycle
4.1 Overview of the Retail Cycle
4.2 Using AI in Planning
4.3 Using AI in Buying
4.4 Using AI in Allocation
4.5 Using AI in In-Season Pricing
4.6 Using AI in Sales and Fulfillment
- Examples of Friendly and Unfriendly AI Problems
- Designing an AI Product for Optimal Performance
- The Role of Measurement in AI Product Success
- Conclusion
Building a Successful AI Product for Retail
In the rapidly evolving world of retail, the use of AI and machine learning has become increasingly important in decision-making processes. As the Director of Scaled Analytics at Nike, I have gained extensive experience in building successful AI products for the retail industry. In this article, I will share with You a framework for building a successful AI product for retail, consisting of three key steps: finding a friendly AI problem, designing the AI product and implementation, and the importance of measurement.
Finding a Friendly AI Problem
Not all AI problems are created equal. In order to build a successful AI product, it is crucial to identify a problem that is compatible with AI and can generate desirable outcomes for the company. The first step in this process is to understand where in the retail cycle AI can be most effectively utilized. The retail cycle consists of various stages, including planning, buying, allocation, in-season pricing, sales, and fulfillment. While AI can be implemented in all these stages, some stages are more friendly to AI than others.
For instance, planning, which involves high-level decisions about revenue, margin, and product breakdown, usually takes place 4-18 months in advance and requires forecasting. However, the accuracy of forecasting decreases the further out the predictions are made. On the other HAND, in-season pricing decisions, which happen closer to the actual sale, can be more accurately forecasted and implemented by AI. Therefore, the later stages of the retail cycle, such as in-season pricing and sales, are more compatible with AI implementation.
Another aspect to consider when finding a friendly AI problem is the lead time before sales and the level of granularity required. Short lead times and low granularity problems are more friendly to AI implementation, as the accuracy of forecasts is higher in these situations. It is important to strike a balance between lead time and granularity to ensure the AI product is effectively meeting the company's objectives.
Designing the AI Product and Implementation
Once an AI problem has been identified, the next step is to design the AI product and the implementation process. The AI product should consist of historical data, forecasting capabilities, optimization strategies, user interfaces, and enforcement mechanisms. Historical data provides the foundation for forecasting future outcomes, while optimization ensures that resources are allocated efficiently to achieve desired objectives. User interfaces and enforcement mechanisms help facilitate the interaction between users and the AI product, ensuring seamless decision-making processes.
However, it is crucial to Align the AI product with the existing business processes and workflows. Failure to do so may result in a lack of adoption and effectiveness of the AI product. It is important to evaluate and potentially redesign processes to effectively leverage the capabilities of the AI product. By integrating the AI product into the existing processes, companies can maximize the benefits of AI while ensuring compatibility with their business operations.
The Importance of Measurement
One often overlooked aspect of building a successful AI product is measurement. It is essential to design a measurement strategy before going live with the AI product. Measurement allows companies to evaluate the effectiveness and impact of the AI product, as well as monitor its performance over time. By measuring key performance indicators (KPIs) and analyzing the data, companies can determine the success of the AI product and make informed decisions for continuous improvement.
Measurement should be aligned with the optimization objectives of the AI product and should consider adoption rates, business constraints, and overall business KPIs. Designing a robust data infrastructure that monitors the performance of the AI product and its impact on business outcomes is crucial for successful measurement. By continuously monitoring and refining the AI product, companies can ensure its long-term success and drive Meaningful results.
Conclusion
Building a successful AI product for the retail industry requires careful consideration of the problem, design, and measurement. By identifying friendly AI problems, designing AI products that align with the existing business processes, and implementing robust measurement strategies, companies can leverage the power of AI to make informed decisions and drive business success in the dynamic world of retail. It is essential to continually iterate and refine the AI product to adapt to evolving business needs and ensure its long-term effectiveness.